1

I have 4 models in the hierarchical regression output. The models are as follows: Model 1: Control variable Model 2: Control variable and 11 independent variables. Model 3: Control variable, 11 independent variables and moderator. Model 4: Control variable, 11 independent variables, moderator and interaction terms.

My question is this:

Which of the 4 models do I consider for final output interpretation? Whether there is a significant relation between different predictors and the outcome variable changes from model to model, in that the relationships that are significant in Model 1 become insignificant in subsequent models.

Alexis
  • 29,850
Muzi
  • 101
  • "Moderator" generally means an independent variable that interacts with another to affect the dependent variable. Therefore, if model 3 has a moderator, it needs an interaction term. I don't see why you would have a model 3 separate from your model 4. – rolando2 Jun 03 '12 at 18:16
  • Thanks for your reply. I added the moderator separately in Model 3 because I wanted to see how it affects the outcome variable when introduced alone. Surprisingly, I found a significant relationship btw the moderator and the outcome variable, but no significant relation between the interaction terms and the outcome variable. I hope I'm doing it the right way. I just wanted to post the result here, but don't know how to do it, please leme know if I can. Thanks for your help:) – Muzi Jun 03 '12 at 19:27

1 Answers1

1

Opinions vary on this, but my view is that you report the model that makes the most substantive sense; the one that advances knowledge the most, answers your research questions the best and so on.

Of course, that presupposes sufficient N to avoid overfitting the model.

You also may want to report all four models; from what you say, it seems like that would add the most information.

Peter Flom
  • 119,535
  • 36
  • 175
  • 383
  • Thanks for your answer. Just wondering if I could post the result here to have your comments. Thanks:) – Muzi Jun 03 '12 at 19:28
  • Sure, why not? And other people may giver their comments too. – Peter Flom Jun 03 '12 at 20:26
  • Hi: Here is the link to the file containing the output ( http://www.mediafire.com/view/?208z1tw155e6juy ). The R square is quite high - is it "too good" for humanities research? Appreciate your views and thank you. – Muzi Jun 03 '12 at 21:03
  • I don't see any reason, offhand, that it's too high. Did you check for high leverage/influential points? Did you plot things? – Peter Flom Jun 04 '12 at 11:22